AbstentionBench - answer unknown - KUQ/Future unknowns - Recall: leaderboard
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama-3.1-Tulu-3-70B-DPO | 81.6 |
| 2 | Llama-3.1-Tulu-3-70B | 80.06 |
| 3 | Gemini 1.5 Pro | 75.77 |
| 4 | Llama-3.1-Tulu-3-8B-DPO | 73.62 |
| 5 | Llama-3.1-Tulu-3-8B | 73.31 |
| 6 | GPT-4o | 71.17 |
| 7 | Llama 3.1 8B Instruct | 69.94 |
| 8 | Llama 3.3 70B Instruct | 69.94 |
| 9 | Mistral 7B Instruct (v0.3) | 68.4 |
| 10 | Qwen 2.5 32B Instruct | 66.56 |
| 11 | O1 | 66.56 |
| 12 | Llama 3.1 405B Instruct | 65.34 |
| 13 | Llama-3.1-Tulu-3-8B-SFT | 62.58 |
| 14 | Llama 3.1 70B Instruct | 62.27 |
| 15 | DeepSeek R1 Distill Llama 70B | 54.29 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-answer-unknown-kuq-future-unknowns-recall · How It Works · Data refreshed daily, snapshot 2026-09-19.